dnn(onnx): test data for quantized layers support pr #28543

[shahkarnav115-beep]

The pr has 6 files with one mobilenetv2-12-qdq.onnx, it was added to check if now qdq model runs on new engine or not.

Related: opencv/opencv#28543

[asmorkalov]

1. Please use existing image from OpenCV Extra as test image. No need to add a new one.
2. Please add the code you used to generate ONNX models with https://github.com/opencv/opencv_extra/blob/4.x/testdata/dnn/onnx/generate_quantized_onnx_models.py. Or, better, reuse existing ONNX files.

[shahkarnav115-beep]

@asmorkalov I am working on the code part to generate the given .onnx files, I will commit the changes shortly.

[shahkarnav115-beep]

@asmorkalov I have added code to generate the 4 specific models (single_qgemm, single_qlinearadd, single_qlinearconv, single_qpool). The mobilenetv2-12-qdq.onn was downloaded from huggingface,

link: https://huggingface.co/onnxmodelzoo/mobilenetv2-12/tree/main

[asmorkalov]

No need to put imagenet_classes.txt into repo. Please remove classes file loading and add assertion with class index check. There are several classification tests in OpenCV, please use the same approach for result validation as other tests.

[asmorkalov]

Please do not put mobilenetv2-12-qdq.onnx into repo, but extend https://github.com/opencv/opencv_extra/blob/4.x/testdata/dnn/download_models.py#L1234 with proper model link. Please use the original HuggingFace link, but not own forks or copies.